Improving Regions of Interest Multivariate Curve Resolution: Development of an Empirical Metric System through the...

The huge potential of liquid chromatography-high-resolution mass spectrometry (LC-HRMS) still comes along with the challenges of data analysis. Regions of interest multivariate curve resolution (ROIMCR) is a valid chemometric tool when working in data-independent acquisition (DIA), since it provides a link between precursor and product ions based on chromatographic and spectral profiles. Still, the quality of the ROIMCR models should be carefully evaluated for a consequent reliable annotation of non-target chemicals. The present case study deals with the non-target analysis of extracts coming from passive samplers deployed in a wastewater treatment facility in Antarctica (Italian Research Station). The extracts, derived from polar organic chemical integrative samplers (POCIS), were analyzed by LC-DIA-HRMS/MS, resulting in a rich and complex data set. The use of a fit-for-purpose ROIMCR workflow ended in six models for a total of 770 resolved components; among them, approximately 100 compounds were tentatively identified thanks to the recently developed MSident software, including pharmaceuticals and natural substances. The chemical meaningfulness of all resolved MCR components was carefully checked and rationalized for the first time in a classification system, with 7 classes divided into 3 "goodness levels" (A, B, and C). Level A components were characterized by single chromatographic peaks and mass spectra with a reasonable appearance of precursor and product ions. Level B components presented flaws or anomalies in either the chromatographic or spectral profile, and level C components clearly showed unacceptable features. The percentage of high-quality MCR components (level A) ranged from 15 to 48%, while components of acceptable quality (levels A and B) reached percentages between 65% and 85%. Most annotated compounds were indeed associated with good-quality MCR components. The automatization of the proposed classification system may constitute a powerful additional tool to evaluate MCR models' quality and thus improve the reliability of ROIMCR results when applied to challenging case studies.